OpenAI GPT-5.6 Sol, Terra & Luna: Complete Guide to the June 2026 Limited Preview

• max reasoning effort: gives the model extra time to work through difficult problems
• ultra multi-subagent mode: uses sub-agents for complex, multi-step work

In its "Ultra" configuration, Sol leads Terminal-Bench 2.1 at 91.9%, a benchmark for complex command-line and agentic workflows . On broader frontier tasks, it benchmarks neck-and-neck with Anthropic's flagship Mythos 5 .

• Pricing: $5 per million input tokens, $30 per million output tokens
• Positioning: For ambitious agentic work where correctness justifies higher latency and cost

Terra — The Balanced Workhorse

Terra is the practical middle tier, designed for everyday production workloads that do not need frontier-level reasoning . OpenAI says Terra delivers "competitive performance to GPT-5.5 while being 2x cheaper" .

• Pricing: $2.50 per million input tokens, $15 per million output tokens
• Positioning: The default model for most production use

Luna — Fast & Low-Cost

Luna is the fastest and most cost-efficient tier, bringing GPT-5.6-level capability at OpenAI's lowest price point . It is built for high-volume, lower-latency workflows while still carrying a "High" capability classification under OpenAI's Preparedness Framework .

• Pricing: $1 per million input tokens, $6 per million output tokens
• Positioning: Volume-tier tasks that need speed and low cost

All three tiers include a 90% discount on cached input tokens and a new 30-minute prompt cache window .

Access Restrictions & U.S. Government Coordination

OpenAI previewed the models' capabilities and shared its rollout plans with the U.S. government ahead of the June 26 launch . At the government's request, OpenAI limited the initial preview to approximately 20 government-approved "trusted partner" organizations, each requiring individual government sign-off . Access is available only through the API and Codex — not in ChatGPT .

OpenAI says this is the first time the U.S. government has preemptively asked an American AI company to restrict a model launch before public release . The company publicly stated that restrictions like this should not become the norm for frontier model releases . Broader access, including ChatGPT, is promised "in the coming weeks" . All three models are classified as "High capability" in both Cybersecurity and CBRN under OpenAI's Preparedness Framework .

Cryptocurrency Community Reaction

The model names triggered immediate and sharp reactions across crypto social media and trading platforms :

• Sol / Solana overlap: The flagship name "Sol" directly matches the ticker symbol for Solana (SOL), one of the largest blockchain networks by market cap. Solana's official X account posted a reply jokingly calling OpenAI CEO Sam Altman "Sam Altcoinman," which quickly went viral . Crypto traders viewed the naming as accidental positive attention for the Solana brand .

• Terra & Luna — the collapsed ecosystem: "Terra" and "Luna" are the names of the blockchain ecosystem that catastrophically collapsed in May 2022, wiping out roughly $40–60 billion in value, including the algorithmic stablecoin UST. The naming brought back painful memories in the crypto community, with some calling it "tone-deaf" or "unfortunate" .

• Market impact: The naming controversy spurred trading FOMO and speculation around related tokens, despite OpenAI explicitly stating the names are strictly for model identification and have no connection to any digital asset projects . Some analysts noted the three-tier naming structure gave the narrative more staying power than a single model name would have .

• Social media buzz: Crypto-focused X accounts and news outlets immediately covered the angle, with posts highlighting that OpenAI "accidentally" picked names loaded with crypto history .

What Happens Next

OpenAI has committed to making GPT-5.6 Sol, Terra, and Luna generally available in ChatGPT, Codex, and the API "in the coming weeks" . For now, access is tightly controlled by government-approved partnerships, marking what may become a precedent for how frontier AI models are deployed in the future.